Let’s start with the headline.
At a major tech event in Taiwan last week, Nvidia (NASDAQ: NVDA) CEO Jensen Huang stood up in front of an audience and called Marvell Technology (NASDAQ: MRVL) the next trillion-dollar company. Not a company with a path to a trillion dollars. The next trillion-dollar company.
Marvell’s stock surged 22% on that comment alone, adding roughly $30 billion in market cap in a single session.
But here’s what’s even more interesting than the endorsement: the numbers Marvell just reported already support it. And there are three massive growth drivers that aren’t even in those numbers yet.
First, What Does Marvell Actually Do?
If you’ve never heard of Marvell, that’s okay. Most people haven’t. It doesn’t make consumer products. It doesn’t have a famous CEO. It doesn’t run ads during the Super Bowl.
What it does is build the chips that make AI data centers work together.
Think of Nvidia’s GPUs as the engines of AI. Blazing fast, incredibly powerful. But engines need roads. They need the wiring, the connectors, the switches that move data between chips, between servers, and between buildings at speeds so fast most people can’t conceptualize them.
That’s Marvell. It builds the connectivity layer — the infrastructure that lets AI actually function at scale.
And right now, that connectivity layer is about to become the biggest bottleneck in AI infrastructure. CEO Matt Murphy said it directly on the earnings call: “In the early stages of generative AI, the primary focus was on addressing compute and memory bottlenecks. As more complex architectures have begun to deploy, the role of networking has become significantly more important.”
Translation: everybody already has enough GPUs. Now they need what connects them.
The Guidance Story Is Unlike Anything On Wall Street Right Now
Here is what Marvell has done to its revenue guidance over the past four quarters. Each time it reports, it has raised the forecast for both the current year AND the year after:
September 2025: FY2027 guidance set at $10 billion. FY2028 at $13 billion. December 2025: FY2027 raised to $11 billion. FY2028 raised to $15 billion. March 2026: FY2027 raised to $11 billion. FY2028 raised to $15 billion. May 2026 (last week): FY2027 raised to $11.5 billion. FY2028 raised to $16.5 billion.
That’s $1.5 billion added to the 2028 outlook in a single quarter. For context, $1.5 billion in annual revenue would be a large company on its own.
Continue Reading With Our 7-Day Free Trial
And the growth rate is accelerating. Data center revenue grew 46% in fiscal 2026. It’s on pace for 50% in fiscal 2027. And Murphy is guiding for 55% growth in fiscal 2028.
It is not normal for a company’s growth rate to accelerate as its revenue base gets bigger. That’s what makes this different.
The Nvidia Partnership Changes Everything
Back in March, Nvidia made a $2 billion investment in Marvell and announced a formal partnership called NVLink Fusion.
Here’s why this matters. Until now, hyperscalers like Amazon, Google, and Microsoft had two separate AI infrastructure choices: buy everything from Nvidia, or build their own custom chips. These two worlds didn’t talk to each other.
NVLink Fusion changes that. It means Marvell’s custom chips can now seamlessly interface with Nvidia’s infrastructure in the same data center — giving hyperscalers full flexibility to mix and match. And Marvell is the bridge that makes that possible.
At the Evercore conference yesterday, Marvell CFO Willem Meintjes put it plainly: hyperscalers “don’t have to choose one or the other. They can choose both.”
That’s not a feature. That’s a structural position in the AI economy.
Three Things That Aren’t In the Numbers Yet
This is the part most investors are missing.
One: Scale-up optics. When AI clusters start spanning multiple buildings and eventually multiple campuses, you need optical connections — light-based data transfer instead of copper wires — to handle the speed and distance. Marvell’s Celestial AI acquisition (closed February 2026) puts it at the center of this. Revenue from scale-up optics is not meaningfully included in the $16.5 billion forecast. Management called the $300 million initial estimate “just the very beginning.” At the Evercore conference, they said it’s targeting a $500 million quarterly run rate by end of fiscal 2028 — growing to $1 billion quarterly by fiscal 2029. That alone is a business the size of a mid-cap semiconductor company.
Two: Scale-up switching. As AI clusters get larger, they need dedicated switches designed specifically for the intense, low-latency traffic AI generates. Marvell’s new 51.2T switch portfolio is just starting to ramp. Murphy said explicitly on the call: “All the numbers I’ve rattled off — scale-up switching is not in any of them. It’s all in front of us.” A business with multibillion-dollar potential that is literally zero in today’s guidance.
Three: Agentic AI. Every time you ask an AI agent to complete a task, it doesn’t just make one request to a model. It might make dozens — routing requests to different parts of the cluster, storing context in memory, calling tools repeatedly. That multiplies the data traffic required. More traffic means more demand for Marvell’s interconnects, switches, and networking chips. Management confirmed: “Agentic AI is not incorporated into our current revenue forecast.” It’s the demand tailwind that hasn’t yet shown up in the numbers.
Who This Hurts & Who It Helps
Broadcom (NASDAQ: AVGO) is Marvell’s most direct competitor in custom silicon and networking. Both companies are winning in this cycle. But Marvell’s NVLink Fusion partnership with Nvidia gives it a structural advantage Broadcom doesn’t have — the ability to serve hyperscalers running both NVIDIA infrastructure and custom chips simultaneously.
The hyperscalers — Amazon (NASDAQ: AMZN), Google (NASDAQ: GOOGL), Microsoft (NASDAQ: MSFT), Meta (NASDAQ: META) — are the customers. Every dollar they spend scaling their AI infrastructure is a dollar that flows, in part, through Marvell’s chips.
Interestingly, both Dell (DELL) and HPE (HPE) — which we covered in detail over the past week — are seeing demand that validates Marvell’s thesis from the buyer side. When Dell says it booked $24.4 billion in AI server orders in a single quarter, that’s demand for infrastructure that runs on exactly the kind of chips Marvell makes.
The Bottom Line
Marvell is not a story about what might happen. It’s a story about what’s already happening — with three large additional growth drivers that haven’t yet been counted.
The guidance has gone up every quarter. The growth rate is accelerating. Jensen Huang endorsed it publicly. Nvidia invested $2 billion in it. And scale-up optics, scale-up switching, and agentic AI tailwinds are all sitting outside the current forecast, waiting to be folded in.
Whether the trillion-dollar market cap arrives in two years or five depends on execution. But the direction of travel has rarely been this clearly signposted in the semiconductor industry.
The stock has more than doubled this year. The guidance trajectory says that’s not the end of the story.
Disclaimer: We do not hold any positions in the above stock(s). Read our full disclaimer here.
